Re: pg_basebackup from 9.4-bdr to 9.4 results in corrupt index

On 5/25/19 11:49 AM, Jānis Pūris wrote:
> Hello,
>
> I'm working with a 9.4-bdr cluster and want to move away from BDR tech all
> together. So my idea was to follow instructions on
> http://bdr-project.org/docs/stable/ to first strip the node from BDR
> making it into "regular" node and then moving the data from this node to
> official 9.4 instance.
>
> Environment:
>
> BDR_Node_1
> BDR_Node_2
>
> Regular_Node_1
>
> Steps done on BDR_Node_2
> 1. Part node from BDR cluster
> http://bdr-project.org/docs/stable/node-management-removing.html
> 2. Remove BDR from node
> http://bdr-project.org/docs/stable/node-management-disabling.html
> 3. Create SR slot select
> pg_create_physical_replication_slot('new_cluster_node');
>
> After this. Steps done on Regular_Node_1, which is 9.4.22
> 1. Init the new node with pg_basebackup with source being BDR_Node_1
> 2. Update recovery.conf and postgres.conf
> 3. Start it with pg_ctl
>
> All seems to be working well, but for this:
>
> select * from information_schema.sequences;
>
> produces "[XX000] ERROR: cache lookup failed for index 899807" for all of
> the DBs that were configured with BDR with various OID.
>
> I can not reproduce this error, when running query on any of the DBs on
> 9.4-bdr node.
>
> Does anyone have any idea what may be causing this ?
> Due to this I'm also unable to run pg_dump on new_cluster_node - it also
> results in the same error.
>
> That said. For more context - If it worked, next steps would have been to
> 1. Deactivate virtual IP on BDR_Node_2
> 2. Check for no replication lag on BDR_Node_2 slot "new_cluster_node"
> 3. Promote Regular_Node_1
> 4. Activate virtual IP on BDR_Node_2
>
> Build the new 9.4 cluster from there on.
>
> Any ideas / comments will be much appreciated!

1. Are you sure that you removed all BDR from the node?
2. Is the corruption there in BDR_Node_1?
3. Can you rebuild the indexes?

--
Angular momentum makes the world go 'round.